
Description
Senior Large Language Model (LLM) Engineer - Fully Remote - VC-backed AI startup
Company Description
Foxy AI is a cutting-edge AI platform that allows influencers, celebrities, and content creators to make AI twins of themselves that their fans can interact with 24 / 7 in any language.
Your Tasks:
We are looking for an experienced and driven Large Language Model (LLM) Engineer to join our dynamic, fully remote international team.
- Your primary focus will be on building datasets and training open-source LLMs (Llama, Mistral, etc.) to provide highly realistic, immersive chat experiences, and assist with their deployment at scale to production.
- You may additionally work on special-purpose ML models and their deployment, in support of our production LLMs and AI imaging systems.
Our Tools & Methodologies:
- Open source LLMs : Llama 3; Mistral; others
- Code : Python 3.10+, with typing; HF Transformers, HF Datasets, Pydantic, PyTorch, NumPy
- Infrastructure : Linux; Docker; AWS; Kubernetes; general cloud GPU providers
- LLM training : Axolotl
- LLM inference : Hugging Face Text Generation Inference (TGI)
- Tools : GitHub
- Ways of Working : Asynchronous text-based communication (Slack) with regular voice / video meetings; 100% remote.
Your Profile:
- Communication : High-quality written English communication and comprehension skills.
- Experience : Minimum 1 year of experience with training open-source LLMs for custom use-cases.
- Ideally : experience of training chat-based / conversational LLMs with an emphasis on highest-quality character realism and best user experience.
- Ideally : experience of training LLMs for adult / NSFW content.
- Open-source LLM : good knowledge of the open-source LLM scene, and up-to-date with current trends in best performing models and practices.
- Good familiarity with the latest models, both foundational and community fine-tunings.
- LLM training : Very experienced with SFT and LORA / qLORA. Some experience with DPO / ORPO / RLHF and other advanced training methods.
- Ideally : experience with, and good understanding of, the Axolotl LLM training framework.
- Dataset creation : Experienced with building, curating and managing datasets for LLM training.
- Especially : experience of using top-quality LLMs (eg GPT-4) for building a dataset to train a smaller open-source LLM, achieving that same quality on a specific task.
- Coding : Good knowledge of Python 3.x, ideally with typing.
- Capable of producing high-quality, zero-error code following Python best practices.
- Ideally : a good understanding of Python typing, plus data validation (Pydantic).
- Python libraries : good familiarity with HF Transformers and HF Datasets.
- Ideally : Experienced with PyTorch, NumPy, Pandas.
- Inference : Ideally experience of LLM inference in production at scale, using inference servers such as HF TGI and / or vLLM.
- ML : Ideally experience of general ML development and training; building custom models to solve specific tasks with an emphasis on being resource-efficient and low-latency.
- Ideally : training both LLMs and Zero-shot learners, lightweight LLMs, CPU code optimization, quantization, code benchmarking, using alpaca-eval metrics to derive / evaluate LLMs.
- Availability : Within the CET / CEST timezone.
What We Offer?
- Impact : A crucial role in building cutting-edge, highly realistic AI characters using open-source LLMs.
- Growth : Opportunities for continuous learning, regular challenges, and knowledge exchange to enhance your skills.
- Culture : A fully remote, flexible, environment in a dynamic, fast-growing company.
- Team : Work with a world-class, ambitious, and supportive team.
- Engagement : Participate in team and company events, as well as regular get-togethers
- Reward : Competitive package consisting of salary and equity including stock options so that you can share the successes that lie ahead.
If you are ready for a new challenge and believe you have the skills and experience we are looking for, we would love to hear from you